https://github.com/agentd00nut/markov_wine

simple markov chain wine reviewer
https://github.com/agentd00nut/markov_wine

Last synced: 1 day ago
JSON representation

simple markov chain wine reviewer

• Host: GitHub
• URL: https://github.com/agentd00nut/markov_wine
• Owner: agentd00nut
• Created: 2017-01-18T17:07:08.000Z (almost 8 years ago)
• Default Branch: master
• Last Pushed: 2017-01-18T17:59:17.000Z (almost 8 years ago)
• Last Synced: 2023-07-21T09:03:59.926Z (over 1 year ago)
• Language: OpenEdge ABL
• Size: 890 KB
• Stars: 0
• Watchers: 3
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# markov_wine

95% of the code was used from here http://www.onthelambda.com/2014/02/20/how-to-fake-a-sophisticated-knowledge-of-wine-with-markov-chains/

All I've done is updated the xpath selector.  Munged around with the process.sh script and tried to improve the review generation.

How to use

==========

As of /1/18/17 the wine-reviews.txt and chain.p are up to date.  This means you can just run `python build_review.py` to generate a review.

If you want to update the library run `./process.sh`... Try to be nice to winespectator.com and either add a sleep in the get-reviews script or

don't do this very often!

Improvements

============

All I've done is reduced the chance that the word "and" is a valid choice more than once in any given review.

Each time "and" is picked it's chances of being used in the review again are halved.

This avoids sentences like `Full-bodied, with fine focus and its broad, intense and vivid, with a tangy, lip-smacking profile.`

Amazingly this tiny modification seems to have improved the results drastically.

Todo

======

Analyze, in some way, a large volume of the generated reviews to identify other stop words (and, or, the, of), that appear

more frequently than one would expect from a real sentence.

Get other data sets and see where it takes us.

Potentially add a "negation" effect on certain key words.  We shouldn't use "sweet" if we've already picked "dry", the blog post touched on this.